First, group this month's duplicates by the entry point that created the second record
Do not start with the merge queue. Start with a report of every duplicate pair found in the last month, and for each pair the creation date and creating source of the newer record: the form it came from, the import batch, the integration that wrote it, or the user who typed it. The CRM fundamentals guide makes the general point that a reporting gap in the CRM is almost always a capture gap at an entry point, and that the report grouping the missing or wrong values by their creating form or import is the one that finds it. Duplicates are the same kind of gap. A record was created when it should have been matched, and the creating source of the newer record says which entry point failed to match.
Group the pairs by that source and rank the groups. The Pareto principle is the right lens, because the entry points are rarely equal: one rebuilt form, one monthly spreadsheet or one integration usually accounts for most of the queue, and the rest is a long tail of hand-typed records and genuine second addresses. The ranking says which door to close first, and it also says when to stop; a group that creates a handful of pairs a month may not be worth a workflow. The framework page is clear that the split is a pattern to check rather than a proportion to expect, so rank the real groups rather than assuming the shape.
Three readings from that report save time before the table. Pairs whose newer record was created on the same day as a form change point at the form, and a rebuilt form is also the commonest reason a routing or field-update rule stops matching, which the diagnosis for a CRM workflow not running starts from. Pairs whose newer records share a creation date and a batch tag point at an import. Pairs whose newer record was created by the integration user, rather than by a form or a person, point at the sync. The data quality guide applies the same discipline to event data: fix the cause that created the bad rows, never the report that showed them.
One case is out of scope here and is worth sending on. If the symptom arrived as a person receiving the same email twice, the diagnosis for subscribers receiving duplicate emails first confirms that it really is one message sent twice to one address, because two systems sending, a flow letting a person in twice and a campaign sent twice all look like duplicate records from the inbox and are not. Duplicate records are one of its six causes; this page is where that cause is traced.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Pairs whose newer record came from one form, often from the day that form changed | A test submission of the form with an address the CRM already holds | A form or lead form creates a new record on every submission instead of matching on email | Set the form to update the matched record; make create-anyway a deliberate choice |
| Pairs whose newer records share a creation date and a batch tag, or rows with a name and company but no address | The import's matching key and what it did with keyless rows | A list import loaded without a matching key, or with rows that hold no usable key | Match on email, fill empty fields only, send keyless rows to review |
| Pairs whose newer record was created by the integration user, or that exist only in the marketing platform | Which system created each copy and whether a merge crossed the sync | The marketing platform and the CRM each create the record and the sync never merges the pair | Let the CRM merge first and pass the surviving identifier across |
| Two records for one person with different addresses, or contacts attached to the wrong company | The addresses and domains on each record | The same person using a work and a personal address, or a shared domain matched to the wrong company | Hold personal-address records for review; list shared domains and every domain a company uses |
| Two company records for one organisation, or a contact that matched nothing | The spellings of company name and other matching values across the pair | Company names and other values spelled differently, so matching rules never see a match | Standardise into a clean field beside the raw one and match on the clean field |
| The pair was merged in the CRM but still appears in the marketing platform, or the reverse | Whether the merge was made in one system or both, and whether the identifier was passed across | Records merged in one system only, so the other keeps both and splits the history | Merge in the system of record first, pass the identifier, and clean before any migration |
Fix, verify and prevent each cause
A form that creates instead of matching
The fundamentals guide lists the predictable sources of duplicates, and a form that creates a new record on every submission is the one that produces the most pairs in the shortest time. Every return visit, every second download and every re-submitted quote request becomes a new person, and the lifecycle stageA CRM field recording where a contact stands in the relationship, such as subscriber, lead, opportunity or customer., owner and history that the first record earned stay on the first record while the workflows start again on the second. The guide's rule is to match on email for contacts and on website domain for companies at every entry point, and, where the CRM supports matching rules, to set them to update an existing record rather than create a new one, with the exception, create anyway, requiring a deliberate choice.
Fix: set the form, and every lead form and embedded form that writes into the CRM, to match on email and update the existing record; where a form has been rebuilt by an agency or moved to a new form tool, check its mapping rather than assuming the old setting survived.
Verify: submit a test entry from an address the CRM already holds and confirm that the existing record was updated and no new record was created; then read the merge queue a week later for pairs from that form.
Prevent: add the matching behaviour to the test every form change has to pass, and keep the list of forms that write into the CRM beside the field documentation so a new form is checked before it goes live.
An import loaded without a key
The tutorial on importing contact lists treats matching as a decision made before the load, not a setting discovered afterwards. Its order is to clean the file to one person per row and remove duplicates inside the file, choose the matching key, decide field by field what a match may change, and send rows with no usable key, such as a name and a company but no email address, to a review list rather than creating records that will become deduplicationRemoving double counts when one conversion is reported by more than one source. work later. The safe default on a match is that the import fills empty fields and adds the person to the new campaign but never overwrites original source, created date, lifecycle stage, owner or consent.
Fix: find the records the import created, which the batch tag makes possible, merge the ones that match an existing person by the written merge rules, and work the keyless rows by hand or discard them. If the list came from an event or a partner, give the surviving records the real source rather than a generic import value.
Verify: load a small sample containing known existing contacts through the corrected rule and confirm that it updates their records and creates none.
Prevent: run every list through the tutorial's steps, with the sample load before the full load and a batch tag on every row, so that a bad import can be found and reversed and the next one cannot repeat it.
Two systems each creating the person
A marketing platform that creates its own record for a contact the CRM already holds, or a CRM that creates one for a subscriber the marketing platform already knows, produces a pair that no form or import explains. The guide to syncing the CRM and the marketing platform names the symptom directly: two records for the same person exist in the marketing platform because the CRM merged them and the marketing side did not hear about it, or because each system created the person on its own path and the sync never recognised them as one. Its rule is that the CRM, as the system of record, merges first and passes the surviving record's identifier to the marketing platform, with a decision in advance about the losing record's engagement history; otherwise the marketing platform keeps two records, sends to both and reports half the engagement on each.
Fix: read which system created each copy of a sample of pairs, write the merge hand-off into the sync contract, and merge the existing pairs in the CRM first with the identifier passed across.
Verify: merge one pair in the CRM and confirm after the next sync that the marketing platform holds one record with the combined history and that the suppressed state, where either copy carried one, survived; a field on the survivor that then keeps changing back after each sync is the diagnosis for CRM values changing back after a sync's subject.
Prevent: write the rule for which records sync and which system may create a record, keep integrations running as their own named users so the audit log shows which system created a copy, and review the sync error queue for duplicate-identifier errors, which are the pairs the sync could not resolve.
Two addresses, or a shared domain
Some pairs are not a capture fault. A person who signs up from a personal address and later from a work address cannot be matched on email at the moment the second record is created, and the guide to matching contacts to company records treats both cases with their own rules. A personal address says nothing reliable about the employer, so that record stays unmatched, in a review queue with a named owner, until evidence arrives; a later sign-up from a work address is one of the forms that evidence takes, and merging the two is part of resolving the case. Shared domains run the other way: a university, a group or an agency domain used by several organisations puts all of them on one company record if matching trusts the domain alone, and a company that uses several domains spawns a new record for each unless every known domain is held on the one record.
Fix: list the domains the business knows to be shared and send their contacts to review; add every known domain to each company record; and deduplicate companies before matching contacts into them, because matching into duplicate companies spreads the problem rather than solving it.
Verify: a test sign-up from a second domain of a known company attaches to the existing record, and a test sign-up from a listed shared domain lands in review rather than on the wrong company.
Prevent: make the domain list and the shared-domain list part of the field documentation, and give the personal-address queue an owner and a cadence so genuine two-address pairs are merged while the people involved are still remembered.
Values spelled so differently that matching never sees a match
Matching rules read a handful of fields, and when those fields hold the same thing under several spellings the rules see different things. The tutorial on standardising CRM field values lists company name among the fields that drift, with legal suffixes, abbreviations, trading names and capitalisation, and names account matching and deduplication as what that drift breaks. The fundamentals guide adds the oldest version of the fault: company names typed with and without the legal suffix. The standardising tutorial's method is to keep the raw value, write a standard value to its own clean field from a mapping table built from the values already in the CRM, send what the table cannot map to a review queue, and fix the spelling at the forms the business controls with a picklistA CRM field type that accepts one value from a defined list of permitted values, so that records can be reported on instead of typed in many spellings.. Accuracy is one of the GDPR's principles, as that tutorial and the site's glossary state it, which is the reason the clean value sits beside the raw one rather than replacing what the person typed.
Fix: build the mapping table for company name and for any other field the matching rules read, point the matching rule at the clean field, and backfill existing records in batches so that pairs the rule could not see become visible in the merge queue.
Verify: create a test company under a new spelling of an existing name and confirm that the clean field maps it and the matching rule attaches it to the existing record.
Prevent: give the review queue an owner who extends the mapping table on a schedule, and trace each new spelling back to the source that produced it, because a source that keeps producing new spellings is better fixed at the form than in the table.
Merged in one system only
A merge made in the CRM while the marketing platform keeps both records, or a merge made in the marketing platform that the CRM never hears of, leaves the pair alive on one side and splits the history between them. The syncing guide owns the day-to-day case and it is handled above; the tutorial on moving to a new CRM owns the version that does the most lasting damage. Its instruction is to clean and deduplicate in the old system first, using the merge rules from the fundamentals guide, because deduplication done after the move works on records that have already lost their activity links, a merge is usually irreversible, and any duplicate loaded into the new system starts creating its own history straight away. The same tutorial loads suppression entries before contacts so that nobody suppressed in the old system is created as reachable in the new one.
Fix: find the pairs that exist in one system only by comparing the identifiers each side holds, merge them in the system of record by the written rules, and pass the surviving identifier across so the other system can merge its copy.
Verify: the address appears once in both systems, the activities and deals of both records sit on the survivor, and a suppressed state on either copy is still present on it.
Prevent: treat a migration's deduplication as a step that happens before the first test load, never after cutover, and keep the merge hand-off in the sync contract so that no merge is ever made on one side alone.
Prevention: matching at every entry point and a merge queue with one owner
Every cause above is a door left open, and the prevention is to put a matching rule on each door and a named person on the queue that catches what the rules miss. The fundamentals guide's instruction, to match on email for contacts and on website domain for companies at every entry point, covers forms, imports, integrations and manual creation, and the table below is that instruction written as a checklist with an owner for each row.
| Entry point | Matching rule | What goes to review | Owner |
|---|---|---|---|
| Web forms, lead forms and embedded forms | Match on email; update the existing record; create anyway only by deliberate choice | Nothing; a form that cannot match is fixed, not reviewed | Whoever owns the forms |
| List imports and bulk updates | Match on email; fill empty fields only; batch tag on every row | Rows with no usable key | Whoever runs the import |
| The marketing platform sync | Only agreed records cross; the CRM merges first and passes the identifier | Duplicate-identifier errors in the sync error queue | Whoever owns the sync |
| Manual creation by sales | Search before creating; company matched on domain | Personal-address contacts awaiting evidence | The personal-address queue owner |
| Company matching | Every known domain on the one record; shared domains listed | Contacts from shared domains | Whoever owns the company structure |
| Migrations and system changes | Deduplicate in the old system before the first test load | Nothing; pairs are resolved before the move | The migration owner |
The merge queue is the second line, and its rules are written before the first merge: the older record's original source wins, the newer record's most recent source wins, the most recent consent value wins, activities and deals are combined, and the owner is the one with the open deal or else the older owner. The fundamentals guide asks for a weekly pass over the queue by a named person rather than an automatic clear, because a merge is irreversible, and a weekly cadence keeps the pairs cheap to resolve while the people involved remember the records. Consent is the rule that matters most. A person who objected to marketing on one record and is reachable on its duplicate has not had their objection honoured, and the ICO's direct marketing guidance, as the site's guides state it, expects an opt-out to be applied in every tool and channel it covers. So the suppressed state wins every merge, the consent recordThe stored evidence of a marketing permission: the channel, the wording the person saw, the date and the source of each opt-in, kept with the contact record. with its source and date travels to the survivor, and the email list hygiene procedure keeps opted-out contacts on the suppression list indefinitely so that a merge never makes someone reachable again.
Then watch the queue the way the data quality guide watches event data: by cause, not by count. A monthly report of new duplicate pairs grouped by the creating source of the newer record is the single most useful check, because a group that grows names the door that reopened. Pairs from one form after a release, pairs from one import batch, pairs created by the integration user, and pairs that are genuine two-address cases each have a different owner from the table above, and the report should route them rather than pile them up. A small, steady number of two-address and shared-domain pairs is the queue doing its job; a sudden jump is a change somebody made that week, and the change log, the form history and the import log should name it.
Finally, keep the structure that makes matching possible. The deduplication entry describes the same discipline for conversions reported by more than one source, and the CRM version has the same shape: one key, applied at every point where a record can be created. The matching guide's company structure, with parents, subsidiaries and a domain list on each record, is what lets contact matching find the right company; the standardising tutorial's clean fields are what let the matching rules read one spelling; and the sync contract is what lets a merge made in one system reach the other. The matching, import, sync and migration guides behind those rules sit together in the CRM section, and the troubleshooting hub routes a symptom that could belong to a record, a workflow or the sync to the page that diagnoses it.
A stair lift installer finds three doors its duplicates come through (illustrative)
Tarnquillo Stairlifts
Tarnquillo Stairlifts, an invented installer of domestic stair lifts, clears a merge queue of about 340 pairs a month and finds it refilled by the following Friday. The operations manager pulls every pair created in the previous 30 days, 352 of them, and groups them by the creating source of the newer record. 211 came from the quote form, which the web agency rebuilt 9 weeks earlier without the matching setting the old form carried, so every returning enquirer became a new person. 83 came from one import: a referral partner's monthly spreadsheet of 400 or so homeowners, loaded without an email key because the partner's file put the address in a notes column. 41 were created by the integration user, because the marketing platform created its own record for contacts that arrived through its landing pages before the sync told the CRM. The remaining 17 were people who had enquired from a personal address and later from a work address.
The installer sets the quote form to match on email and update the existing record, sends the partner's keyless rows to a review list and asks the partner for a file with the address in its own column, and rewrites the sync so that landing-page contacts are created in the CRM first and the CRM merges before the identifier crosses. The 352 pairs are merged by written rules over two weeks, with the most recent consent value winning and 6 suppressed records staying suppressed. The following month the queue holds 14 new pairs, all two-address cases, and the monthly report by creating source shows no pair from the quote form, the import or the integration user.
Common mistakes
- Clearing the queue without grouping it. The pairs go and the doors stay open; next week's queue is this week's queue with new dates.
- Merging before the rules are written. Which original source, which owner and which consent value survive is decided in the moment, differently each time, and cannot be undone.
- Deleting the newer record. The opt-out, the open deal or the sequence in progress may sit on it; merge by rules instead.
- Trusting a rebuilt form to keep its matching setting. An agency template or a new form tool starts from its own defaults.
- Loading a list because it looked clean. A file with no usable key creates a record for every row; sample first and tag the batch.
- Merging in the marketing platform alone. The CRM keeps both records and the next sync recreates the pair.
- Guessing a company from a personal address or a shared domain. A wrong match is worse than no match; hold it for review.
- Treating two-address pairs as a fault to engineer away. They are the merge queue's proper work; the fault is everything else in it.
